Expand group dataset metadata field for slice media types

fiftyone/core/dataset.py

@@ -5,6 +5,7 @@
 | `voxel51.com <https://voxel51.com/>`_
 |
 """
+
 from collections import defaultdict
 import contextlib
 from datetime import datetime
@@ -6588,6 +6589,20 @@ def _expand_schema(self, samples, dynamic):
                         field_name, value.name, sample.media_type
                     )
 
+                if (
+                    self.media_type == fom.GROUP
+                    and sample.media_type
+                    not in set(self.group_media_types.values())
+                ):
+                    expanded |= self._sample_doc_cls.merge_field_schema(
+                        {
+                            "metadata": fo.EmbeddedDocumentField(
+                                fome.get_metadata_cls(sample.media_type)
+                            )
+                        },
+                        validate=False,
+                    )
+
                 if not dynamic and field_name in schema:
                     continue
 
@@ -8548,7 +8563,7 @@ def _finalize_frames(sample_collection, key_field, frame_key_field):
 
 
 def _get_media_type(sample):
-    for field, value in sample.iter_fields():
+    for _, value in sample.iter_fields():
         if isinstance(value, fog.Group):
             return fom.GROUP
 

fiftyone/core/metadata.py

@@ -5,6 +5,7 @@
 | `voxel51.com <https://voxel51.com/>`_
 |
 """
+
 import itertools
 import logging
 import multiprocessing.dummy
@@ -229,6 +230,23 @@ def compute_sample_metadata(sample, overwrite=False, skip_failures=False):
         sample.save()
 
 
+def get_metadata_cls(media_type):
+    """Get the ``metadata`` class for a media_type
+
+    Args:
+        media_type (str): a media type value
+
+    Returns:
+        a :class:`Metadata` class
+    """
+    if media_type == fom.IMAGE:
+        return ImageMetadata
+    elif media_type == fom.VIDEO:
+        return VideoMetadata
+
+    return Metadata
+
+
 def compute_metadata(
     sample_collection,
     overwrite=False,

tests/unittests/group_tests.py

@@ -5,6 +5,7 @@
 | `voxel51.com <https://voxel51.com/>`_
 |
 """
+
 from itertools import groupby
 import json
 import os
@@ -21,6 +22,7 @@
 import fiftyone.utils.data as foud
 import fiftyone.utils.groups as foug
 import fiftyone.core.media as fom
+import fiftyone.core.metadata as fome
 from fiftyone import ViewExpression as E, ViewField as F
 
 from decorators import drop_datasets
@@ -2103,6 +2105,36 @@ def test_group_by_clips(self):
         self.assertTrue(also_view.is_saved)
         self.assertEqual(len(also_view), 2)
 
+    @drop_datasets
+    def test_expand_group_metadata(self):
+        dataset: fo.Dataset = fo.Dataset()
+
+        group = fo.Group()
+        samples = [
+            fo.Sample(filepath="video.mp4", group=group.element("video")),
+            fo.Sample(filepath="image.png", group=group.element("image")),
+        ]
+        dataset.add_samples(samples)
+
+        # assert that slices have their media type metadata fields populated
+        for (
+            name,
+            field,
+        ) in fome.ImageMetadata._fields.items():  # pylint: disable=no-member
+            self.assertIsInstance(
+                dataset.get_field(f"metadata.{name}", include_private=True),
+                field.__class__,
+            )
+
+        for (
+            name,
+            field,
+        ) in fome.VideoMetadata._fields.items():  # pylint: disable=no-member
+            self.assertIsInstance(
+                dataset.get_field(f"metadata.{name}", include_private=True),
+                field.__class__,
+            )
+
 
 def _make_group_by_dataset():
     sample_id1 = ObjectId()